उत्तर_समतुल्यता

  • विवरण :

उत्तर तुल्यता डेटासेट में SQuAD डेटासेट पर कई मॉडलों से मॉडल भविष्यवाणियों पर मानव रेटिंग शामिल हैं। रेटिंग यह स्थापित करती हैं कि अनुमानित उत्तर सोने के उत्तर के 'समतुल्य' है (प्रश्न और संदर्भ दोनों को ध्यान में रखते हुए)।

अधिक विशेष रूप से, 'समतुल्य' से हमारा मतलब है कि अनुमानित उत्तर में कम से कम वही जानकारी होती है जो सोने के उत्तर में होती है और इसमें अनावश्यक जानकारी नहीं होती है। डेटासेट में निम्नलिखित के लिए एनोटेशन शामिल हैं: * SQuAD देव पर BiDAF से भविष्यवाणियां * SQuAD देव पर XLNet से भविष्यवाणियां * SQuAD देव पर ल्यूक से भविष्यवाणियां * SQuAD प्रशिक्षण, देव और परीक्षण उदाहरणों पर अल्बर्ट से भविष्यवाणियां

विभाजित करना उदाहरण
'ae_dev' 4,446
'ae_test' 9,724
'dev_bidaf' 7,522
'dev_luke' 4,590
'dev_xlnet' 7,932
'train' 9,090
  • फ़ीचर संरचना :
FeaturesDict({
    'candidate': Text(shape=(), dtype=string),
    'context': Text(shape=(), dtype=string),
    'gold_index': int32,
    'qid': Text(shape=(), dtype=string),
    'question': Text(shape=(), dtype=string),
    'question_1': ClassLabel(shape=(), dtype=int64, num_classes=3),
    'question_2': ClassLabel(shape=(), dtype=int64, num_classes=3),
    'question_3': ClassLabel(shape=(), dtype=int64, num_classes=3),
    'question_4': ClassLabel(shape=(), dtype=int64, num_classes=3),
    'reference': Text(shape=(), dtype=string),
    'score': float32,
})
  • फ़ीचर दस्तावेज़ीकरण :
विशेषता कक्षा आकार डीटाइप विवरण
विशेषताएं डिक्ट
उम्मीदवार मूलपाठ डोरी
संदर्भ मूलपाठ डोरी
gold_index टेन्सर int32
qid मूलपाठ डोरी
प्रश्न मूलपाठ डोरी
प्रश्न 1 क्लासलेबल int64
प्रश्न 2 क्लासलेबल int64
प्रश्न 3 क्लासलेबल int64
प्रश्न_4 क्लासलेबल int64
संदर्भ मूलपाठ डोरी
अंक टेन्सर फ्लोट32
  • उद्धरण :
@article{bulian-etal-2022-tomayto,
      title={Tomayto, Tomahto. Beyond Token-level Answer Equivalence for Question Answering Evaluation},
      author={Jannis Bulian and Christian Buck and Wojciech Gajewski and Benjamin Boerschinger and Tal Schuster},
      year={2022},
      eprint={2202.07654},
      archivePrefix={arXiv},
      primaryClass={cs.CL}
}